Remote sensing image water body change detection method and device, electronic equipment and storage medium

Abstract

The present invention provides a remote sensing image water body change detection method and device, electronic equipment and storage medium, and relates to the field of remote sensing image processing technology. The method comprises: obtaining a first remote sensing image and a second remote sensing image obtained by shooting a first area with water bodies at different times, taking the two as target images, respectively, performing multi-scale feature extraction on the two, and performing feature enhancement processing on the extraction results, performing feature map acquisition processing and feature map fusion processing on the results of the feature enhancement processing, obtaining a feature map of the water body image enhancement feature at each scale and a fusion feature map of the target image, obtaining an aggregate feature map and a spatial attention feature map of the target image, and determining the water body change area in the first remote sensing image and the second remote sensing image based on the two and a threshold nonlinear mapping model and outputting the corresponding first image. The present invention can improve the detection result accuracy and detection efficiency of remote sensing image water body change detection.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a remote sensing image water body change detection method and device, electronic equipment and storage medium. Background Art

[0002] Remote sensing image water change detection refers to the process of identifying water body differences in remote sensing images of the same object area obtained at different times. It is widely used in urban water area monitoring, land water loss, flood disaster assessment and ecological environment monitoring.

[0003] At present, due to the complex content and rich semantics of water body images in remote sensing images, and the influence of factors such as illumination changes, ecological environment and mismatch errors in actual situations, changes in real water areas are difficult to detect, and the accuracy of detection results and efficiency of the detection process of water body change detection in remote sensing images are both low. Summary of the invention

[0004] The present invention provides a remote sensing image water body change detection method and device, electronic equipment and storage medium, which are used to solve the defects of low detection result accuracy and low detection process efficiency in remote sensing image water body change detection in the prior art, and achieve the purpose of improving the detection result accuracy and detection efficiency of remote sensing image water body change detection.

[0005] The present invention provides a remote sensing image water body change detection method, comprising the following steps.

[0006] A first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time are obtained; wherein, water bodies are distributed in the first area; the first remote sensing image and the second remote sensing image are respectively used as target images, and based on a feature enhancement network model, multi-scale feature extraction is performed on the target image and feature enhancement processing is performed on the result of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales; feature map acquisition processing and feature map fusion processing are performed on the water body image enhancement features at multiple scales based on a feature map acquisition and fusion network model to obtain feature maps corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, and the resolution of the feature map at the first scale is greater than that of the feature map at other scales in the multiple scales. The resolution of the feature map of the second scale is smaller than the resolution of the feature maps of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output.

[0007] According to a remote sensing image water body change detection method provided by the present invention, after outputting a first image of a water body change area, the method further includes: marking the area range of the water body change area in the first remote sensing image or the second remote sensing image according to the water body change area; and outputting a second image of the marked area range.

[0008] According to a remote sensing image water body change detection method provided by the present invention, before obtaining a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time, the method also includes: obtaining multiple groups of remote sensing image samples; wherein each group of remote sensing image samples in the multiple groups of remote sensing image samples includes a first remote sensing sample image of a second area shot at a third time and a second remote sensing sample image of the second area shot at a fourth time, and water bodies are distributed in the second area; marking water body areas in the first remote sensing sample image and the second remote sensing sample image to obtain at least one water body area label; wherein the types of water body area labels include: changed and unchanged; according to the multiple groups of remote sensing image samples and at least one water body area label of each group of remote sensing image samples, training multiple initial neural network models to obtain multiple target neural network models corresponding to the multiple initial neural network models one by one; wherein the multiple target neural network models include: a feature enhancement network model, a feature map acquisition and fusion network model, a feature aggregation network model and a spatial attention network model.

[0009] According to a remote sensing image water body change detection method provided by the present invention, multiple initial neural network models are trained, including: training the multiple initial neural network models based on a preset loss function; wherein the loss function is obtained based on a weighted sum of a first loss function, a second loss function and a third loss function; the first loss function is determined according to a first pixel number, a second pixel number and a third pixel number, the first pixel number is the number of pixels that are the same in the predicted change information and the water body area label in at least one group of remote sensing image samples input this time during the training of the multiple initial neural network models, and the second pixel number is the number of pixels that are the same in the predicted change information and the water body area label in at least one group of remote sensing image samples input this time, and the second pixel number is the number of pixels that are the same in the water body area label ... The third pixel number is the number of pixels whose water area labels are changed in at least one group of remote sensing image samples input this time; the second loss function is determined according to the binary mapping parameter values ​​of the predicted change information in at least one group of remote sensing image samples input this time and the binary mapping parameter values ​​of the water area labels in at least one group of remote sensing image samples input this time; the third loss function is determined according to the number of pixel pairs that have not changed, the number of pixel pairs that have changed, the height of the remote sensing image samples, and the width and distance mapping of the remote sensing image samples in at least one group of remote sensing image samples input this time.

[0010] According to a remote sensing image water body change detection method provided by the present invention, a spatial attention feature map of a first remote sensing image and a spatial attention feature map of a second remote sensing image are segmented based on a convolutional network model to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image, including: segmenting the spatial attention feature map of the first remote sensing image into multiple first segmented images based on the convolutional network model, and segmenting the spatial attention feature map of the second remote sensing image into multiple second segmented images based on the convolutional network model; wherein the multiple first segmented images correspond one-to-one to the multiple second segmented images; calculating the similarity between the target first segmented image and its corresponding target second segmented image; when the similarity is greater than or equal to a similarity threshold, determining that the predicted change information of each pixel in the target first segmented image and the target second segmented image is unchanged; when the similarity is less than the similarity threshold, determining that the predicted change information of each pixel in the target first segmented image and the target second segmented image is changed.

[0011] According to a remote sensing image water body change detection method provided by the present invention, the step length of cropping in multi-scale feature extraction is 50px.

[0012] The present invention also provides a remote sensing image water body change detection device, comprising the following modules:

[0013] An acquisition module is used to acquire a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time; wherein water bodies are distributed in the first area;

[0014] A first processing module is used to use the first remote sensing image and the second remote sensing image as target images respectively, perform multi-scale feature extraction on the target images based on a feature enhancement network model, and perform feature enhancement processing on the results of the multi-scale feature extraction to obtain enhanced features of water body images at multiple scales;

[0015] A second processing module is used to perform feature map acquisition processing and feature map fusion processing on the water body image enhancement features at multiple scales based on the feature map acquisition and fusion network model, obtain a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, the resolution of the feature map at the first scale is greater than the resolution of the feature maps at other scales in the multiple scales, and the resolution of the feature map at the second scale is less than the resolution of the feature maps at other scales in the multiple scales;

[0016] A third processing module is used to aggregate the fused feature maps at multiple scales based on a feature aggregation network model to obtain an aggregated feature map of the target image;

[0017] A fourth processing module, for obtaining a spatial attention feature map of the target image based on the spatial attention network model and the aggregated feature map;

[0018] A prediction module is used to segment the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image based on a convolutional network model to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change;

[0019] A determination module, for performing threshold nonlinear mapping on the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information based on a threshold nonlinear mapping model, and determining a water body change area in the first remote sensing image and the second remote sensing image;

[0020] The output module is used to output the first image of the water body change area.

[0021] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the remote sensing image water body change detection method described in any one of the above methods is implemented.

[0022] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the remote sensing image water body change detection method as described in any one of the above is implemented.

[0023] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for detecting water body changes in remote sensing images.

[0024] The present invention provides a method and device for detecting water body changes in remote sensing images, an electronic device, and a storage medium. The method comprises acquiring a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein water bodies are distributed in the first area; taking the first remote sensing image and the second remote sensing image as target images respectively, and performing multi-scale feature extraction on the target image based on a feature enhancement network model and performing feature enhancement processing on the result of the multi-scale feature extraction to obtain enhanced features of water body images at multiple scales; performing feature map acquisition processing and feature map fusion processing on the enhanced features of water body images at multiple scales based on a feature map acquisition and fusion network model to obtain feature maps corresponding to the enhanced features of water body images at each scale in the multiple scales, and obtaining a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, and the feature map at the first scale The resolution of the feature map of the first remote sensing image is greater than the resolution of the feature map of other scales in the multiple scales, and the resolution of the feature map of the second scale is less than the resolution of the feature map of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output. It can be seen that the embodiment of the present invention can solve the defects of low detection result accuracy and detection process efficiency of remote sensing image water body change detection in the prior art, and achieve the purpose of improving the detection result accuracy and detection efficiency of remote sensing image water body change detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 This is one of the flow charts of the method for detecting water changes in remote sensing images provided by an embodiment of the present invention.

[0027] Figure 2 This is an example diagram of a specific process in the remote sensing image water body change detection method provided by an embodiment of the present invention.

[0028] Figure 3 This is the second flow chart of the method for detecting water changes in remote sensing images provided by an embodiment of the present invention.

[0029] Figure 4 It is a diagram of the input remote sensing image and the output processing effect in the remote sensing image water body change detection method provided by an embodiment of the present invention.

[0030] Figure 5 It is a schematic diagram of the structure of a remote sensing image water body change detection device provided by an embodiment of the present invention.

[0031] Figure 6 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] In the related art, due to the influence of many factors, the real changes in the real water area are difficult to detect in two remote sensing images taken at different times. The specific reasons are as follows: First, because the water images in remote sensing images are relatively complex, semantically rich and densely distributed, the appearances such as the shape and color of the water bodies at different times are usually quite different. Therefore, the detection of water changes cannot be simply located and detected directly by the physical appearance of the water images; at the same time, due to the influence of illumination changes, ecological environment and mismatch errors, the changes in pseudo information such as noise generated lack robustness, making it difficult to detect the real changes in the water area; secondly, because the water area is diverse and has a large span in scale, in this case, the use of a single scale feature extraction cannot meet the requirements of obtaining effective information related to changes from the water image. In this case, when obtaining the deep features of the water area, redundant information will be reused many times, and the representation ability of the global and local details of the context will be ignored, resulting in serious loss of target detail information, and the accuracy and efficiency of the detection results are low. In addition, when selecting detection samples, the importance of sample category balance is ignored. For example, the sample images of the selected changed class are much smaller than the sample images of the unchanged class, which leads to a serious category imbalance problem and ultimately reduces the accuracy of change detection.

[0034] Based on the above-mentioned problems, an embodiment of the present invention provides a method for detecting water body changes in remote sensing images, so as to effectively improve the detection result accuracy and detection efficiency of water body change detection in remote sensing images.

[0035] Combine the following Figure 1-Figure 4 The present invention describes the method for detecting water body changes in remote sensing images.

[0036] Figure 1 This is one of the flow charts of the remote sensing image water body change detection method provided by the present invention. Figure 1 As shown, the method includes the following S110~S180.

[0037] S110: Acquire a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time; wherein water bodies are distributed in the first area.

[0038] The first time and the second time are different times. The second time is after the first time.

[0039] S120: The first remote sensing image and the second remote sensing image are respectively used as target images, and based on the feature enhancement network model, multi-scale feature extraction is performed on the target images and feature enhancement processing is performed on the results of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales.

[0040] When performing multi-scale feature extraction on a target image, the target image can be cropped based on a preset cropping step size to obtain cropped images of multiple scales, thereby obtaining water body image features of multiple scales.

[0041] When performing multi-scale feature extraction, the target image can be cropped or expanded into multiple cropped images of different scales based on a preset cropping step size.

[0042] For example, the cropping step size is 50px, based on which the target image is cropped or expanded into a plurality of cropped images of different scales, such as 350px×350px, 400px×400px, 450px×450px, and 500px×500px.

[0043] It is understandable that the above-listed cropping step sizes are merely exemplary, and in the embodiment of the present invention, the cropping step size includes but is not limited to 50px, and may also be 60px, 65px, and so on.

[0044] When performing multi-scale feature extraction on a target image, multi-scale feature extraction can be performed on the target image based on a Swin-tansformer model.

[0045] For example, the first remote sensing image is Xa , based on the Swin-tansformer model for X a By performing multi-scale feature extraction, we can obtain the first remote sensing image X a Corresponding water image features at multiple scales , , and .

[0046] , , and Obtained by the following operations: .in, Represents the processing operation of the Swin-tansformer model, s represents the stage corresponding to the scale, for example, s=0, the corresponding scale is scale 1; s=1, the corresponding scale is scale 2; s=3, the corresponding scale is scale 3; s=4, the corresponding scale is scale 4.

[0047] For the above scale 1, scale 2, scale 3 and scale 4, the size relationship is as follows: scale 1 < scale 2 < scale 3 < scale 4. For example, scale 1, scale 2, scale 3 and scale 4 can be 350px×350px, 400px×400px, 450px×450px, 500px×500px respectively.

[0048] It should be noted that the specific values ​​of scale 1, scale 2, scale 3 and scale 4 listed above are merely exemplary. The specific values ​​of scale 1, scale 2, scale 3 and scale 4 include but are not limited to the values ​​listed above, as long as they meet the multiple scales obtained based on the preset cropping step size.

[0049] Similarly, the second remote sensing image is X b , based on the Swin-tansformer model for X b By performing multi-scale feature extraction, the second remote sensing image X can be obtained. b Corresponding water image features at multiple scales , , and .in, , , and Obtained by the following operations: .

[0050] When performing feature enhancement processing on the results of multi-scale feature extraction, the water image features at multiple scales can be squeezed based on the convolution operation, and a model can be created using the local attention mechanism. The corresponding data can be processed based on the model to refine the water image features at multiple scales while reducing redundant information, thereby achieving the purpose of highlighting the significant features of the water image features at multiple scales.

[0051] For example, for the first remote sensing image X a , get the water body image features Corresponding water image enhancement features The operation process is as follows: ,in, express The convolution operation, express The convolution operation, Represents the local attention operation, which is implemented by the model created using the local attention mechanism.

[0052] For example, for the second remote sensing image X b , get the water body image features Corresponding water image enhancement features The operation process is as follows: .

[0053] In some embodiments, Figure 2 As shown, the feature enhancement network model is Figure 2 Medium FEM, Feature Enhanced Network Model , , is a remote sensing image set to be detected, the remote sensing image set includes multiple groups of remote sensing images, each group of remote sensing images includes two remote sensing images taken of the same area at different times, such as the first remote sensing image and the second remote sensing image mentioned above. Represents a batch of remote sensing images, H represents the length of the remote sensing image, and W represents the width of the remote sensing image.

[0054] S130: Based on the feature map acquisition and fusion network model, feature map acquisition processing and feature map fusion processing are performed on the water body image enhancement features at multiple scales to obtain a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on the feature map pyramid network FPN, the resolution of the feature map at the first scale is greater than the resolution of the feature maps at other scales in the multiple scales, and the resolution of the feature map at the second scale is less than the resolution of the feature maps at other scales in the multiple scales.

[0055] In the feature map acquisition and fusion network model, feature map acquisition processing can be first performed on the water body image enhancement features at multiple scales based on the Feature Pyramid Networks (FPN), and the feature map corresponding to the water body image enhancement features at each scale in the multiple scales can be obtained.

[0056] For example, for the first remote sensing image X a , in obtaining water body image features Corresponding water image enhancement features In this case, obtain The corresponding feature map The operation process is as follows: ,in, represents the operation of the feature map pyramid network. In this case, the feature map , feature map , feature map and feature map .

[0057] For example, for the second remote sensing image X b , in obtaining water body image features Corresponding water image enhancement features In this case, obtain The corresponding feature map The operation process is as follows: In this case, the feature map can be obtained , feature map , feature map and feature map .

[0058] It should be noted that the above , The downsampling rate is relatively large, which makes the features after global average pooling focus more on high-level semantics.

[0059] After that, based on the feature map obtained above, the feature map at the first scale with the largest resolution and the feature map at the second scale with the smallest resolution are fused to obtain a fused feature map of the target image. Obtaining the fused feature map in the above manner is conducive to forming a complementary relationship between high-level features and low-level features, and reducing feature deviation.

[0060] For example, a first remote sensing image X is obtained. a Fusion feature map The specific operation process is as follows: = ;in, represents the upsampling operation, represents the downsampling operation, represents feature concatenation, Based on the first remote sensing image X a The feature map obtained at the first scale, Based on the first remote sensing image X a The feature map obtained at the second scale.

[0061] For example, the second remote sensing image X is obtained b Fusion feature map The specific operation process is as follows: = ;in, Based on the second remote sensing image X b The feature map obtained at the first scale, Based on the second remote sensing image X b The feature map obtained at the second scale.

[0062] S140: Based on the feature aggregation network model, the fused feature maps at the above multiple scales are aggregated to obtain an aggregated feature map of the target image.

[0063] For example, for the first remote sensing image X a , get the first remote sensing image X a Aggregate feature map of The operation process is as follows: ;in, Represents a max-pooling operation.

[0064] For example, for the second remote sensing image X b , obtain the second remote sensing image X b Aggregate feature map of The operation process is as follows: .

[0065] In some embodiments, Figure 2 As shown in the figure, the operations of feature map acquisition and fusion network model and feature aggregation network model are integrated into one model, and the following is obtained: Figure 2 UPPFPN in.

[0066] S150: Based on the spatial attention network model and the aggregated feature map, a spatial attention feature map of the target image is obtained.

[0067] For example, for the first remote sensing image X a , the aggregate feature map obtained above Input spatial attention network model, based on the spatial attention network model Perform spatial attention operation to obtain the spatial attention feature map output by the spatial attention network model The specific operation process is as follows: ;in, Represents the spatial attention operation.

[0068] For example, for the second remote sensing image X b , the aggregate feature map obtained above Input spatial attention network model, based on the spatial attention network model Perform spatial attention operation to obtain the spatial attention feature map output by the spatial attention network model The specific operation process is as follows: .

[0069] In some embodiments, Figure 2 As shown in Figure 2, the spatial attention network model is as follows: Figure 2 SAM in.

[0070] S160: Segment the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image based on the convolutional network model to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change.

[0071] For example, the operation process of obtaining the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image is as follows: ;in, represents matrix multiplication, Indicates forecast change information.

[0072] In some embodiments, the spatial attention feature map of the first remote sensing image is segmented into multiple first segmented images based on a convolutional network model, and the spatial attention feature map of the second remote sensing image is segmented into multiple second segmented images based on a convolutional network model; wherein the multiple first segmented images correspond one-to-one to the multiple second segmented images; the similarity between the target first segmented image and its corresponding target second segmented image is calculated; when the similarity is greater than or equal to a similarity threshold, the predicted change information of each pixel in the target first segmented image and the target second segmented image is determined to be unchanged; when the similarity is less than a similarity threshold, the predicted change information of each pixel in the target first segmented image and the target second segmented image is determined to be changed.

[0073] In some embodiments, Figure 2 As shown in Figure 2 FFM in.

[0074] S170: Based on the threshold nonlinear mapping model, threshold nonlinear mapping is performed on the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image, and the predicted change information to determine the water body change area in the first remote sensing image and the second remote sensing image.

[0075] The first remote sensing image X a After performing a 3×3 convolution operation on the spatial attention feature map, a 1×1 convolution operation is performed on the 3×3 convolution result to obtain the first processing result . The acquisition process is as follows: .

[0076] The second remote sensing image X b After performing a 3×3 convolution operation on the spatial attention feature map, a 1×1 convolution operation is performed on the 3×3 convolution result to obtain the second processing result. . The acquisition process is as follows: .

[0077] Based on the threshold nonlinear mapping model, the process of obtaining the water body change area in the first remote sensing image and the second remote sensing image is as follows: ;in, Represents a threshold nonlinear operation, such as a softmax nonlinear operation.

[0078] When Y=1, it indicates change, and 0≤Y≤1 indicates no change. Based on the value of Y, the area where the water body changes can be determined.

[0079] In some embodiments, Figure 2 As shown, the threshold nonlinear mapping model is as follows Figure 2 Medium SM.

[0080] S180: Output a first image of the water body change area.

[0081] For example, Figure 2 As shown, the first image 210 is finally output. Among them, Figure 2 The white area in the first image shown is the area where the water body changes.

[0082] In some embodiments, before executing S110, Figure 3 As shown, the following S310~S330 may also be executed.

[0083] S310: Acquire multiple groups of remote sensing image samples; wherein each group of remote sensing image samples in the multiple groups of remote sensing image samples includes a first remote sensing sample image of a second area taken at a third time and a second remote sensing sample image of the second area taken at a fourth time, and there is water body distributed in the second area.

[0084] The remote sensing sample images in the multiple groups of remote sensing image samples can be collected satellite images. In specific implementations, the multiple groups of remote sensing images can be divided into training samples, verification samples and test samples. In addition, several super large images can be collected separately to verify the detection efficiency and accuracy.

[0085] S320: Label the water body areas in the first remote sensing sample image and the second remote sensing sample image to obtain at least one water body area label; wherein the types of water body area labels include: changed and unchanged.

[0086] The labeling can be done manually or using corresponding labeling software to label the same water area parts in the first remote sensing sample image and the second remote sensing sample image as unchanged, and label the different water area parts in the first remote sensing sample image and the second remote sensing sample image as changed.

[0087] S330: Based on multiple groups of remote sensing image samples and at least one water body area label of each group of remote sensing image samples, multiple initial neural network models are trained to obtain multiple target neural network models corresponding to the multiple initial neural network models; wherein the multiple target neural network models include: feature enhancement network model, feature map acquisition and fusion network model, feature aggregation network model and spatial attention network model.

[0088] The multiple initial neural network models include: an initial feature enhancement network model, an initial feature map acquisition and fusion network model, an initial feature aggregation network model and an initial spatial attention network model.

[0089] The initial feature enhancement network model is trained to obtain the above-mentioned feature enhancement network model; the initial feature map acquisition and fusion network model is trained to obtain the above-mentioned feature map acquisition and fusion network model; the initial feature aggregation network model is trained to obtain the above-mentioned feature aggregation network model; the initial spatial attention network model is trained to obtain the spatial attention network model.

[0090] In some embodiments, in order to reduce the deviation of the obtained target neural network model, multiple initial neural network models are trained based on a preset loss function; wherein the loss function is obtained based on a weighted sum of a first loss function, a second loss function, and a third loss function.

[0091] The first loss function is determined based on the first number of pixels, the second number of pixels, and the third number of pixels. The first number of pixels is the number of pixels with the same predicted change information and water area labels in at least one group of remote sensing image samples input this time during the training process, the second number of pixels is the number of pixels whose predicted change information is changed in at least one group of remote sensing image samples input this time, and the third number of pixels is the number of pixels whose water area labels are changed in at least one group of remote sensing image samples input this time.

[0092] The second loss function is determined according to the binary mapping parameter values ​​of the predicted change information in at least one set of remote sensing image samples input this time and the binary mapping parameter values ​​of the water area labels in at least one set of remote sensing image samples input this time.

[0093] The third loss function is determined based on the number of unchanged pixel pairs, the number of changed pixel pairs, the height of the remote sensing image samples, and the width and distance mapping of the remote sensing image samples in at least one group of remote sensing image samples input this time.

[0094] For example, the loss function can be calculated by the following formula: ;in, is the loss function, is the first loss function, is the second loss function, is the third loss function, is the weight coefficient of the first loss function, is the weight coefficient of the second loss function.

[0095] The first loss function It can be calculated by the following formula: ; Wherein, P is the number of pixels whose change information is predicted to be changed, and G is the number of pixels whose change information is predicted to be unchanged. Represents the number of intersection pixels between the predicted change information and the actual water area labels.

[0096] The second loss function It can be calculated by the following formula: ; where y represents the parameter value of the true label, Parameter value representing the predicted change information.

[0097] The third loss function It can be calculated by the following formula:

[0098] ;

[0099] Among them, b is the batch number of the batch processing, i is the height, j is the width, is the number of unchanged pixel pairs, is the number of pixel pairs that change. 1≤b≤B0, 1≤i≤H0, 1≤j≤W0. m=2. , The distance mapping is obtained as follows: the similarity between the first remote sensing sample image and the second remote sensing sample image is calculated, and the similarity is mapped to a corresponding preset value. is a parameter representing similarity, The larger the value, the higher the similarity. The smaller it is, the lower the similarity.

[0100] It can be calculated by the following formula: ;in, is a binary label mapping for a batch.

[0101] It can be calculated by the following formula: .

[0102] In some embodiments, Figure 2 As shown, the parameter values ​​of the parameters obtained in UPPFPN and FFM can be obtained by detecting the head head, based on the loss function The parameter values ​​of the obtained parameters are corrected.

[0103] The loss function designed in the present invention can effectively supervise the network to learn better feature representation, thereby improving the accuracy of change detection in the target area in remote sensing water body images. It is worth noting that compared with the pixels in the surrounding tissue, the areas near the boundaries of the change area contain more detailed information about the cluttered background and even the nearby water body. The internal pixels located in the water change area can better represent the characteristics of the target than the pixels close to the boundary of the change area. Therefore, the feature potential of each pixel is closely related to the shape of the water area, not just the distance from each pixel to the center of the change.

[0104] Based on the embodiment of the present invention, the water body changes in the remote sensing image are detected, and the experimental results are shown in Table 1.

[0105] Table 1

[0106]

[0107] In some embodiments, after outputting the first image of the water body change area, the embodiments of the present invention can also mark the area range of the water body change area in the first remote sensing image or the second remote sensing image according to the water body change area, and then output a second image of the marked area range.

[0108] like Figure 4 As shown, based on Figure 4 The first remote sensing image shown in part a and Figure 4 The second remote sensing image shown in part b in FIG. 1 can be output as follows through the embodiment of the present invention: Figure 4 The first image shown in part c, and the output is Figure 4 The second image is shown in part d. Figure 4 The red marked area in the second image shown in part d is the area range of the water body change area.

[0109] The remote sensing image water body change detection method provided by the present invention is provided by acquiring a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein water bodies are distributed in the first area; the first remote sensing image and the second remote sensing image are respectively used as target images, and based on a feature enhancement network model, multi-scale feature extraction is performed on the target image and feature enhancement processing is performed on the result of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales; feature map acquisition processing and feature map fusion processing are performed on the water body image enhancement features at multiple scales based on a feature map acquisition and fusion network model to obtain feature maps corresponding to the water body image enhancement features at each scale in multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, and the resolution of the feature map at the first scale is greater than that of the multi-scale feature map. The resolution of the feature map of other scales in the scales is smaller than the resolution of the feature map of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output. It can be seen that the embodiment of the present invention can solve the defects of low detection result accuracy and detection process efficiency of remote sensing image water body change detection in the prior art, and achieve the purpose of improving the detection result accuracy and detection efficiency of remote sensing image water body change detection.

[0110] The remote sensing image water body change detection device provided by the present invention is described below. The remote sensing image water body change detection device described below and the remote sensing image water body change detection method described above can be referenced to each other.

[0111] Figure 5 Schematic diagram of the structure of the remote sensing image water body change detection device provided by the embodiment of the present invention. Figure 5 As shown, the remote sensing image water body change detection device 500 includes: an acquisition module 501, a first processing module 502, a second processing module 503, a third processing module 504, a fourth processing module 505, a prediction module 506, a determination module 507 and an output module 508.

[0112] The acquisition module 501 is used to acquire a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time; wherein water bodies are distributed in the first area.

[0113] The first processing module 502 is used to use the first remote sensing image and the second remote sensing image as target images respectively, perform multi-scale feature extraction on the target images based on the feature enhancement network model, and perform feature enhancement processing on the results of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales.

[0114] The second processing module 503 is used to perform feature map acquisition processing and feature map fusion processing on the water body image enhancement features at multiple scales based on the feature map acquisition and fusion network model, obtain the feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on the feature map pyramid network FPN, the resolution of the feature map at the first scale is greater than the resolution of the feature maps at other scales in the multiple scales, and the resolution of the feature map at the second scale is less than the resolution of the feature maps at other scales in the multiple scales.

[0115] The third processing module 504 is used to perform aggregation processing on the fused feature maps at multiple scales based on the feature aggregation network model to obtain an aggregated feature map of the target image.

[0116] The fourth processing module 505 is used to obtain a spatial attention feature map of the target image based on the spatial attention network model and the aggregated feature map.

[0117] The prediction module 506 is used to segment the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image based on the convolutional network model to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change.

[0118] The determination module 507 is used to perform threshold nonlinear mapping on the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information based on the threshold nonlinear mapping model to determine the water body change area in the first remote sensing image and the second remote sensing image.

[0119] The output module 508 is used to output the first image of the water body change area.

[0120] The remote sensing image water body change detection device provided by the present invention obtains a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein water bodies are distributed in the first area; the first remote sensing image and the second remote sensing image are respectively used as target images, and based on a feature enhancement network model, multi-scale feature extraction is performed on the target image and feature enhancement processing is performed on the result of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales; feature map acquisition processing and feature map fusion processing are performed on the water body image enhancement features at multiple scales based on a feature map acquisition and fusion network model to obtain feature maps corresponding to the water body image enhancement features at each scale in multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, and the resolution of the feature map at the first scale is greater than that of the multi-scale feature map. The resolution of the feature map of other scales in the scales is smaller than the resolution of the feature map of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output. It can be seen that the embodiment of the present invention can solve the defects of low detection result accuracy and detection process efficiency of remote sensing image water body change detection in the prior art, and achieve the purpose of improving the detection result accuracy and detection efficiency of remote sensing image water body change detection.

[0121] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor (processor) 610 , a communication interface (Communications Interface) 620 , a memory (memory) 630 and a communication bus 640 , wherein the processor 610 , the communication interface 620 , and the memory 630 communicate with each other through the communication bus 640 . The processor 610 can call the logic instructions in the memory 630 to execute the remote sensing image water body change detection method, which includes: obtaining a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein, there is water distributed in the first area; taking the first remote sensing image and the second remote sensing image as target images respectively, based on the feature enhancement network model, performing multi-scale feature extraction on the target image and performing feature enhancement processing on the result of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales; performing feature map acquisition processing and feature map fusion processing on the water body image enhancement features at multiple scales based on the feature map acquisition and fusion network model, obtaining a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtaining a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on the feature map pyramid network FPN, the first scale The resolution of the feature map of the first scale is greater than the resolution of the feature maps of other scales in the multiple scales, and the resolution of the feature map of the second scale is less than the resolution of the feature maps of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output.

[0122] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0123] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the remote sensing image water body change detection method provided by the above methods, the method including: acquiring a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein, water bodies are distributed in the first area; taking the first remote sensing image and the second remote sensing image as target images respectively, and performing multi-scale feature extraction on the target image based on a feature enhancement network model and performing feature enhancement processing on the result of the multi-scale feature extraction to obtain water body image enhancement features at multiple scales; performing feature map acquisition processing and feature map fusion processing on the water body image enhancement features at multiple scales based on a feature map acquisition and fusion network model, obtaining a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtaining a target image obtained by fusing the feature map at the first scale with the feature map at the second scale. A fused feature map; wherein the feature map is obtained based on a feature map pyramid network FPN, the resolution of the feature map of the first scale is greater than the resolution of the feature maps of other scales in the multiple scales, and the resolution of the feature map of the second scale is less than the resolution of the feature maps of other scales in the multiple scales; based on a feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain an aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, a spatial attention feature map of the target image is obtained; based on a convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on a threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and a first image of the water body change area is output.

[0124] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the remote sensing image water body change detection method provided by the above-mentioned methods, the method comprising: acquiring a first remote sensing image obtained by shooting a first area at a first time and a second remote sensing image obtained by shooting the first area at a second time; wherein, water bodies are distributed in the first area; taking the first remote sensing image and the second remote sensing image as target images respectively, and performing multi-scale feature extraction on the target image and feature enhancement processing on the result of the multi-scale feature extraction based on a feature enhancement network model to obtain water body image enhancement features at multiple scales; performing feature map acquisition processing and feature map fusion processing on the water body image enhancement features at multiple scales based on a feature map acquisition and fusion network model to obtain feature maps corresponding to the water body image enhancement features at each scale in the multiple scales, and obtaining a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is based on The feature map pyramid network FPN obtains that the resolution of the feature map of the first scale is greater than the resolution of the feature map of other scales in the multiple scales, and the resolution of the feature map of the second scale is less than the resolution of the feature map of other scales in the multiple scales; based on the feature aggregation network model, the fused feature maps at multiple scales are aggregated to obtain the aggregated feature map of the target image; based on the spatial attention network model and the aggregated feature map, the spatial attention feature map of the target image is obtained; based on the convolutional network model, the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented to obtain the predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; based on the threshold nonlinear mapping model, the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image and the predicted change information are threshold nonlinearly mapped to determine the water body change area in the first remote sensing image and the second remote sensing image; and the first image of the water body change area is output.

[0125] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0126] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting water changes in remote sensing images, characterized in that: include: Acquire a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time; wherein water bodies are distributed in the first area; The first remote sensing image and the second remote sensing image are respectively used as target images, and based on the feature enhancement network model, multi-scale feature extraction is performed on the target images and feature enhancement processing is performed on the results of the multi-scale feature extraction to obtain enhanced features of water body images at multiple scales; Based on the feature map acquisition and fusion network model, feature map acquisition processing and feature map fusion processing are performed on the water body image enhancement features at the multiple scales to obtain a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, the resolution of the feature map at the first scale is greater than the resolution of the feature maps at other scales in the multiple scales, and the resolution of the feature map at the second scale is less than the resolution of the feature maps at other scales in the multiple scales; Based on a feature aggregation network model, aggregating the fused feature maps at the multiple scales to obtain an aggregated feature map of the target image; Based on the spatial attention network model and the aggregated feature map, obtaining a spatial attention feature map of the target image; Segmenting the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image based on a convolutional network model to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; Based on a threshold nonlinear mapping model, performing threshold nonlinear mapping on the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image, and the predicted change information to determine water body change areas in the first remote sensing image and the second remote sensing image; A first image of the water body change area is output.

2. The method for detecting water changes in remote sensing images according to claim 1, characterized in that: After outputting the first image of the water body change area, the method further includes: According to the water body change area, marking the area range of the water body change area in the first remote sensing image or the second remote sensing image; A second image with the region range marked is output.

3. The method for detecting water changes in remote sensing images according to claim 1, characterized in that: Before acquiring a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time, the method further includes: Acquire multiple groups of remote sensing image samples; wherein each group of remote sensing image samples in the multiple groups of remote sensing image samples includes a first remote sensing sample image of a second area taken at a third time and a second remote sensing sample image of the second area taken at a fourth time, and water bodies are distributed in the second area; Annotating water areas in the first remote sensing sample image and the second remote sensing sample image to obtain at least one water area label; wherein the types of the water area labels include: changed and unchanged; Training a plurality of initial neural network models according to the plurality of groups of remote sensing image samples and the at least one water body region label of each group of remote sensing image samples to obtain a plurality of target neural network models corresponding to the plurality of initial neural network models; Among them, the multiple target neural network models include: the feature enhancement network model, the feature map acquisition and fusion network model, the feature aggregation network model and the spatial attention network model.

4. The method for detecting water changes in remote sensing images according to claim 3, characterized in that: The training of multiple initial neural network models includes: Training the multiple initial neural network models based on a preset loss function; Wherein, the loss function is obtained based on weighted summation of the first loss function, the second loss function and the third loss function; The first loss function is determined according to a first number of pixels, a second number of pixels, and a third number of pixels, wherein the first number of pixels is the number of pixels with the same predicted change information and water body area labels in at least one group of remote sensing image samples input this time during the training of the multiple initial neural network models, the second number of pixels is the number of pixels with predicted change information being changed in the at least one group of remote sensing image samples input this time, and the third number of pixels is the number of pixels with changed water body area labels in the at least one group of remote sensing image samples input this time; The second loss function is determined according to the binary mapping parameter value of the predicted change information in the at least one set of remote sensing image samples input this time and the binary mapping parameter value of the water area label in the at least one set of remote sensing image samples input this time; The third loss function is determined based on the number of unchanged pixel pairs, the number of changed pixel pairs, the height of the remote sensing image samples, and the width and distance mapping of the remote sensing image samples in at least one group of remote sensing image samples input this time.

5. The method for detecting water changes in remote sensing images according to claim 4, characterized in that: The spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image are segmented based on a convolutional network model to obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image, including: Based on a convolutional network model, the spatial attention feature map of the first remote sensing image is segmented into a plurality of first segmented images, and based on the convolutional network model, the spatial attention feature map of the second remote sensing image is segmented into a plurality of second segmented images; wherein the plurality of first segmented images correspond to the plurality of second segmented images one by one; Calculating the similarity between the target first segmented image and its corresponding target second segmented image; In a case where the similarity is greater than or equal to a similarity threshold, determining that the predicted change information of each pixel in the target first segmented image and the target second segmented image is unchanged; When the similarity is less than the similarity threshold, the predicted change information of each pixel in the target first segmented image and the target second segmented image is determined to be a change.

6. The method for detecting water changes in remote sensing images according to claim 1, characterized in that: The step size of cropping in the multi-scale feature extraction is 50px.

7. A remote sensing image water body change detection device, characterized in that: include: An acquisition module, used to acquire a first remote sensing image obtained by photographing a first area at a first time and a second remote sensing image obtained by photographing the first area at a second time; wherein water bodies are distributed in the first area; A first processing module is used to use the first remote sensing image and the second remote sensing image as target images respectively, perform multi-scale feature extraction on the target images based on a feature enhancement network model, and perform feature enhancement processing on the results of the multi-scale feature extraction to obtain enhanced features of water body images at multiple scales; A second processing module is used to perform feature map acquisition processing and feature map fusion processing on the water body image enhancement features at the multiple scales based on the feature map acquisition and fusion network model, obtain a feature map corresponding to the water body image enhancement features at each scale in the multiple scales, and obtain a fused feature map of the target image obtained by fusing the feature map at the first scale and the feature map at the second scale; wherein the feature map is obtained based on a feature map pyramid network FPN, the resolution of the feature map at the first scale is greater than the resolution of the feature maps at other scales in the multiple scales, and the resolution of the feature map at the second scale is less than the resolution of the feature maps at other scales in the multiple scales; A third processing module is used to perform aggregation processing on the fused feature maps at the multiple scales based on a feature aggregation network model to obtain an aggregated feature map of the target image; A fourth processing module, configured to obtain a spatial attention feature map of the target image based on the spatial attention network model and the aggregated feature map; A prediction module, configured to segment the spatial attention feature map of the first remote sensing image and the spatial attention feature map of the second remote sensing image based on a convolutional network model, and obtain predicted change information of each pixel on the water body image in the first remote sensing image and the second remote sensing image; wherein the predicted change information includes: change or no change; A determination module, configured to perform threshold nonlinear mapping on the fused feature map of the first remote sensing image, the fused feature map of the second remote sensing image, and the predicted change information based on a threshold nonlinear mapping model, so as to determine a water body change area in the first remote sensing image and the second remote sensing image; An output module is used to output a first image of the water body change area.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for detecting water body changes in remote sensing images as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting water body changes in remote sensing images as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting water body changes in remote sensing images as described in any one of claims 1 to 6 is implemented.

Images